Can We Predict Stock Prices?

What Does "Predicting" Even Mean?

When people ask whether stock prices can be predicted, they're usually conflating three very different questions — each with a different answer. The failure to separate them leads to some of the most common and costly investing mistakes.

• Will this stock be higher or lower tomorrow? Pure short-term price prediction. This is the domain of high-frequency traders and algo shops with microsecond data feeds.
• Will this stock outperform the market over 3–5 years? Fundamental forecasting based on business quality, valuation, and competitive position. A different animal entirely.
• When will the next crash happen? Macro timing — predicting human behavior at civilizational scale. The domain of endless, often wrong, macroeconomic punditry.

These three questions have meaningfully different answers. Blurring them is how financial media generates clicks and how investors get led astray. Let's be precise.

The Efficient Market Hypothesis — What It Actually Says

In 1970, economist Eugene Fama published his landmark paper formalizing the Efficient Market Hypothesis (EMH): the idea that asset prices fully reflect all available information. If true in its strictest form, predicting prices is by definition impossible — you can't consistently profit from information the market has already priced in.

Fama articulated three forms, each making progressively stronger claims:

• Weak form: Prices reflect all past trading data — price history, volume, patterns. Implication: technical analysis adds no systematic edge.
• Semi-strong form: Prices reflect all publicly available information — earnings reports, news, analyst estimates. Implication: fundamental analysis provides no consistent edge either.
• Strong form: Prices reflect all information, including private/insider information. Implication: no one can consistently beat the market.

The practical truth is nuanced. The strong form is clearly false — insider trading works (which is why it's illegal). The weak form has strong empirical support. The semi-strong form is where the real debate lives, and where active investors place their bets.

The most damning evidence against consistent prediction: over 15-year periods, more than 85% of active fund managers underperform their benchmark index (SPIVA data, S&P Dow Jones Indices). These are professionals with Bloomberg terminals, analyst teams, and decades of experience — and they still can't reliably beat a passive index.

What the Data Actually Shows About Short-Term Prediction

The random walk theory, popularized by Burton Malkiel's A Random Walk Down Wall Street (1973), holds that day-to-day stock price movements are largely random — the result of unpredictable new information arriving continuously. A stock that was up yesterday is not meaningfully more likely to be up tomorrow.

This has been tested exhaustively. The autocorrelation of daily stock returns — how much today's return predicts tomorrow's — is typically near zero for individual large-cap stocks. There is essentially no signal in the noise at daily time scales accessible to retail investors.

Consider the implication: if you could predict daily price direction with just 55% accuracy (barely better than a coin flip), and you used modest leverage, you would be the richest person alive within a few years. No such person exists sustainably. The few who approach this use co-location servers, proprietary order flow data, and strategies that can't scale beyond a few hundred million dollars.

The key distinction is between short-term noise and long-term signal:

What CAN Be Predicted (Partially)

Despite the randomness at short time scales, decades of academic research have identified a handful of genuinely persistent signals — factors that have predicted returns with statistical significance across markets and time periods. These are not magic formulas; they work on average, over long periods, with significant variance. But they are real.

• 1. Valuation mean reversion (strong evidence): Stocks with very high P/E ratios tend to underperform over the next decade relative to cheap stocks. Yale economist Robert Shiller's CAPE ratio (cyclically adjusted P/E) has historically predicted 10-year market returns with meaningful accuracy. Overpaying matters — but only over years, not months.
• 2. Earnings momentum (moderate evidence): Stocks that beat earnings estimates tend to drift higher in subsequent weeks — a phenomenon called "post-earnings announcement drift" (PEAD). The market is slow to fully price in earnings surprises, creating a temporary edge for attentive investors.
• 3. Value factor (historically reliable, compressed): Cheap stocks — low P/B, low P/E — have outperformed over long periods. The premium appears to have compressed since the 1990s as it became widely known and arbitraged.
• 4. Quality factor: Companies with high Return on Invested Capital (ROIC), low debt, and stable or growing earnings consistently tend to outperform lower-quality peers over multi-year periods. Great businesses compound; mediocre businesses revert.
• 5. Price momentum (controversial): Stocks that have risen over 6–12 months tend to continue rising — until they abruptly reverse. Momentum is real but dangerous near turning points and requires disciplined exit rules.

Machine Learning and AI — Has It Solved Prediction?

The most sophisticated attempt to predict stock prices is happening inside quantitative hedge funds. Renaissance Technologies, Two Sigma, D.E. Shaw, and Citadel collectively employ thousands of PhDs in mathematics, physics, and computer science, running models on petabytes of alternative data. If anyone can predict markets, it's them.

Renaissance's Medallion fund is the most striking data point: it returned approximately 66% annualized before fees from 1988 to 2018 — one of the greatest track records in financial history. Prediction is possible. But consider the caveats:

• Medallion has been closed to outside investors for decades. The people running it keep all the alpha.
• It employs 300+ PhDs and has built proprietary data infrastructure over 30+ years.
• Its strategies trade hundreds of thousands of positions at microsecond precision — impossible to replicate.
• The edge constantly erodes as others discover and copy patterns. What worked in 2005 is gone by 2015.

For retail investors and most institutional managers, the picture is far less rosy. ML models trained on historical market data almost always overfit — they find patterns in past data that sound compelling but fail to generalize. The market is a non-stationary system: it changes in response to the very strategies trying to exploit it.

What This Means for How You Should Invest

The evidence points to a clear and liberating conclusion: stop trying to predict next week's stock price, and start focusing on the things that actually drive long-term returns. Warren Buffett captured the core insight decades ago: "In the short run, the market is a voting machine. In the long run, it's a weighing machine."

Short-term prices are driven by sentiment, news flow, and randomness — none of which you can reliably predict. Long-term returns are driven by business quality, earnings compounding, and the price you pay — all of which you can research and assess. Focus there.

The market's short-term randomness isn't just a problem to work around — it's an opportunity. Random sentiment-driven selloffs in high-quality businesses create buying opportunities at temporarily attractive valuations. That's the edge available to patient, fundamental investors.

The practical takeaways are straightforward:

• Don't try to time your entry with precision — focus on valuation and business quality instead
• Buy great businesses at fair or cheap prices and hold them long enough for the weighing machine to work
• Use volatility as your ally: temporary price drops in quality businesses are gifts, not disasters
• Avoid speculative positions in businesses you can't understand or value — randomness works against you there

Use Valuation and Quality Metrics to Find Stocks Worth Owning Long-Term

Forget the noise. BriMindInvest gives you AI-powered scores, ROIC analysis, valuation metrics, and fundamental quality data — the signals that actually matter over a 3–10 year horizon.